Guardrails

Guardrails are an Enterprise-only feature. Contact [email protected] to upgrade.

Guardrails provide content safety, security, and compliance controls for your LLM applications. Detect and block prompt injections, PII, secrets, and custom content violations.

Get Guardrail Configuration

Retrieve the guardrail configuration for an organization.

organizationId

string

required

Organization ID

config

object | null

Guardrail configuration object (null if not configured)

Show Configuration Object

string

Configuration ID

organizationId

string

Organization ID

enabled

boolean

Whether guardrails are enabled

systemRules

object | null

System rule configurations

Show child attributes

prompt_injection

object

Show child attributes

enabled

boolean

Enable detection

action

'block' | 'redact' | 'warn' | 'allow'

Action to take

jailbreak

object

Jailbreak attempt detection

pii_detection

object

PII (Personally Identifiable Information) detection

secrets

object

API keys and secrets detection

file_types

object

File type restrictions

document_leakage

object

Document leakage prevention

maxFileSizeMb

number

Maximum file size in MB

allowedFileTypes

string[]

Allowed MIME types for file uploads

piiAction

'block' | 'redact' | 'warn' | 'allow' | null

Default action for PII detection

createdAt

date

Configuration creation date

updatedAt

date

Last update date

curl https://api.llmgateway.io/guardrails/config/org_abc123 \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN"

{
  "id": "gc_xyz789",
  "organizationId": "org_abc123",
  "enabled": true,
  "systemRules": {
    "prompt_injection": {
      "enabled": true,
      "action": "block"
    },
    "jailbreak": {
      "enabled": true,
      "action": "block"
    },
    "pii_detection": {
      "enabled": true,
      "action": "redact"
    },
    "secrets": {
      "enabled": true,
      "action": "block"
    },
    "file_types": {
      "enabled": true,
      "action": "block"
    },
    "document_leakage": {
      "enabled": false,
      "action": "warn"
    }
  },
  "maxFileSizeMb": 10,
  "allowedFileTypes": [
    "image/jpeg",
    "image/png",
    "image/gif",
    "image/webp"
  ],
  "piiAction": "redact",
  "createdAt": "2024-01-10T10:00:00Z",
  "updatedAt": "2024-01-15T14:30:00Z"
}

Update Guardrail Configuration

Create or update the guardrail configuration.

organizationId

string

required

Organization ID

enabled

boolean

Enable or disable all guardrails

systemRules

object

System rule configurations

Show child attributes

prompt_injection

object

Show child attributes

enabled

boolean

Enable detection

action

'block' | 'redact' | 'warn' | 'allow'

Action

jailbreak

object

Jailbreak detection config

pii_detection

object

PII detection config

secrets

object

Secrets detection config

file_types

object

File type config

document_leakage

object

Document leakage config

maxFileSizeMb

number

Maximum file size in MB

allowedFileTypes

string[]

Array of allowed MIME types

piiAction

'block' | 'redact' | 'warn' | 'allow'

Default PII action

curl -X PUT https://api.llmgateway.io/guardrails/config/org_abc123 \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "enabled": true,
    "systemRules": {
      "prompt_injection": {
        "enabled": true,
        "action": "block"
      },
      "pii_detection": {
        "enabled": true,
        "action": "redact"
      }
    },
    "maxFileSizeMb": 20,
    "piiAction": "redact"
  }'

{
  "id": "gc_xyz789",
  "organizationId": "org_abc123",
  "enabled": true,
  "systemRules": {
    "prompt_injection": {
      "enabled": true,
      "action": "block"
    },
    "jailbreak": {
      "enabled": true,
      "action": "block"
    },
    "pii_detection": {
      "enabled": true,
      "action": "redact"
    },
    "secrets": {
      "enabled": true,
      "action": "block"
    },
    "file_types": {
      "enabled": true,
      "action": "block"
    },
    "document_leakage": {
      "enabled": false,
      "action": "warn"
    }
  },
  "maxFileSizeMb": 20,
  "allowedFileTypes": [
    "image/jpeg",
    "image/png",
    "image/gif",
    "image/webp"
  ],
  "piiAction": "redact",
  "createdAt": "2024-01-10T10:00:00Z",
  "updatedAt": "2024-01-16T09:15:00Z"
}

Reset Configuration

Reset guardrail configuration to defaults.

organizationId

string

required

Organization ID

curl -X POST https://api.llmgateway.io/guardrails/config/org_abc123/reset \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN"

List Custom Rules

Retrieve all custom guardrail rules for an organization.

organizationId

string

required

Organization ID

rules

array

Array of custom rule objects

curl https://api.llmgateway.io/guardrails/rules/org_abc123 \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN"

{
  "rules": [
    {
      "id": "rule_123",
      "organizationId": "org_abc123",
      "name": "Block Competitor Names",
      "type": "blocked_terms",
      "config": {
        "type": "blocked_terms",
        "terms": ["CompetitorA", "CompetitorB"],
        "matchType": "contains",
        "caseSensitive": false
      },
      "priority": 100,
      "enabled": true,
      "action": "block",
      "createdAt": "2024-01-10T10:00:00Z",
      "updatedAt": "2024-01-10T10:00:00Z"
    },
    {
      "id": "rule_456",
      "organizationId": "org_abc123",
      "name": "Credit Card Pattern",
      "type": "custom_regex",
      "config": {
        "type": "custom_regex",
        "pattern": "\\b\\d{4}[\\s-]?\\d{4}[\\s-]?\\d{4}[\\s-]?\\d{4}\\b"
      },
      "priority": 90,
      "enabled": true,
      "action": "redact",
      "createdAt": "2024-01-12T14:30:00Z",
      "updatedAt": "2024-01-12T14:30:00Z"
    }
  ]
}

Create Custom Rule

Create a new custom guardrail rule.

organizationId

string

required

Organization ID

name

string

required

Rule name

type

'blocked_terms' | 'custom_regex' | 'topic_restriction'

required

Rule type

config

object

required

Rule configuration (varies by type)

priority

number

default:100

Rule priority (higher = checked first)

enabled

boolean

default:true

Enable the rule

action

'block' | 'redact' | 'warn' | 'allow'

default:"block"

Action to take on violation

Rule Types

Blocked Terms

Block or redact specific words/phrases:

{
  "type": "blocked_terms",
  "config": {
    "type": "blocked_terms",
    "terms": ["badword1", "badword2"],
    "matchType": "contains",
    "caseSensitive": false
  }
}

Match Types:

exact: Exact word match
contains: Substring match
regex: Regular expression pattern

Custom Regex

Match content using regex patterns:

{
  "type": "custom_regex",
  "config": {
    "type": "custom_regex",
    "pattern": "\\b\\d{3}-\\d{2}-\\d{4}\\b"
  }
}

Topic Restriction

Block or allow specific topics:

{
  "type": "topic_restriction",
  "config": {
    "type": "topic_restriction",
    "blockedTopics": ["violence", "illegal activities"],
    "allowedTopics": ["technology", "education"]
  }
}

curl -X POST https://api.llmgateway.io/guardrails/rules/org_abc123 \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Block Profanity",
    "type": "blocked_terms",
    "config": {
      "type": "blocked_terms",
      "terms": ["badword1", "badword2"],
      "matchType": "contains",
      "caseSensitive": false
    },
    "priority": 100,
    "action": "block"
  }'

{
  "id": "rule_789",
  "organizationId": "org_abc123",
  "name": "Block Profanity",
  "type": "blocked_terms",
  "config": {
    "type": "blocked_terms",
    "terms": ["badword1", "badword2"],
    "matchType": "contains",
    "caseSensitive": false
  },
  "priority": 100,
  "enabled": true,
  "action": "block",
  "createdAt": "2024-01-16T10:00:00Z",
  "updatedAt": "2024-01-16T10:00:00Z"
}

Update Custom Rule

Update an existing custom rule.

organizationId

string

required

Organization ID

ruleId

string

required

Rule ID

curl -X PATCH https://api.llmgateway.io/guardrails/rules/org_abc123/rule_789 \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "enabled": false,
    "priority": 110
  }'

Delete Custom Rule

Delete a custom guardrail rule.

organizationId

string

required

Organization ID

ruleId

string

required

Rule ID

curl -X DELETE https://api.llmgateway.io/guardrails/rules/org_abc123/rule_789 \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN"

{
  "success": true
}

List Violations

Retrieve guardrail violations with filtering and pagination.

organizationId

string

required

Organization ID

cursor

string

Pagination cursor from previous response

limit

number

default:50

Number of results per page (1-100)

startDate

string

Filter by start date (ISO 8601 with timezone)

endDate

string

Filter by end date (ISO 8601 with timezone)

actionTaken

'blocked' | 'redacted' | 'warned'

Filter by action taken

ruleId

string

Filter by specific rule ID

curl "https://api.llmgateway.io/guardrails/violations/org_abc123?limit=50&actionTaken=blocked" \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN"

{
  "violations": [
    {
      "id": "vio_123",
      "organizationId": "org_abc123",
      "logId": "log_456",
      "ruleId": "system:prompt_injection",
      "ruleName": "Prompt Injection Detection",
      "category": "injection",
      "actionTaken": "blocked",
      "matchedPattern": "ignore previous instructions",
      "matchedContent": "Ignore previous instructions and...",
      "contentHash": "abc123def456",
      "apiKeyId": "key_xyz789",
      "model": "gpt-4o",
      "createdAt": "2024-01-16T10:30:00Z"
    }
  ],
  "pagination": {
    "nextCursor": "vio_124",
    "hasMore": true,
    "limit": 50
  }
}

Get Violation Statistics

Retrieve aggregated violation statistics.

organizationId

string

required

Organization ID

days

number

default:7

Number of days to analyze (1-90)

curl "https://api.llmgateway.io/guardrails/stats/org_abc123?days=30" \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN"

{
  "blocked": 45,
  "redacted": 23,
  "warned": 12,
  "total": 80
}

Test Content

Test content against guardrail rules without logging violations.

organizationId

string

required

Organization ID

content

string

required

Content to test

curl -X POST https://api.llmgateway.io/guardrails/test/org_abc123 \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "This is test content with a badword1 in it"
  }'

{
  "passed": false,
  "blocked": true,
  "violations": [
    {
      "ruleId": "rule_789",
      "ruleName": "Block Profanity",
      "category": "blocked_terms",
      "action": "block",
      "matchedPattern": "badword1",
      "matchedContent": "badword1"
    }
  ],
  "rulesChecked": 5
}

List System Rules

Get information about available system rules.

curl https://api.llmgateway.io/guardrails/system-rules \
  -H "Authorization: Bearer YOUR_SESSION_TOKEN"

{
  "rules": [
    {
      "id": "system:prompt_injection",
      "name": "Prompt Injection Detection",
      "category": "injection",
      "defaultEnabled": true,
      "defaultAction": "block"
    },
    {
      "id": "system:jailbreak",
      "name": "Jailbreak Prevention",
      "category": "jailbreak",
      "defaultEnabled": true,
      "defaultAction": "block"
    },
    {
      "id": "system:pii_detection",
      "name": "PII Detection",
      "category": "pii",
      "defaultEnabled": true,
      "defaultAction": "redact"
    },
    {
      "id": "system:secrets",
      "name": "Secrets Detection",
      "category": "secrets",
      "defaultEnabled": true,
      "defaultAction": "block"
    },
    {
      "id": "system:file_types",
      "name": "File Type Restrictions",
      "category": "files",
      "defaultEnabled": true,
      "defaultAction": "block"
    },
    {
      "id": "system:document_leakage",
      "name": "Document Leakage Prevention",
      "category": "document_leakage",
      "defaultEnabled": false,
      "defaultAction": "warn"
    }
  ]
}

Guardrail Actions

Each rule can have one of four actions:

Block

{"action": "block"}

Reject the request entirely. Returns a 400 error. Use for:

Security threats (prompt injection, jailbreaks)
Policy violations
Prohibited content

Redact

{"action": "redact"}

Remove/mask the matched content and continue processing. Use for:

PII (emails, phone numbers, SSNs)
Sensitive data
Credit card numbers

Warn

{"action": "warn"}

Log the violation but allow the request to proceed. Use for:

Monitoring suspicious patterns
Low-risk content
Testing new rules

Allow

{"action": "allow"}

Explicitly allow content (for whitelisting). Use for:

Overriding other rules
Approved exceptions
False positive handling

Error Responses

{
  "message": "Guardrails require an enterprise plan"
}

Best Practices

Start Conservative

Enable core system rules (prompt injection, jailbreak, secrets)
Use “warn” action initially to monitor false positives
Gradually tighten rules based on violation patterns

Layer Your Defense

Combine system rules with custom rules
Use both blocking and redaction
Set appropriate priorities for rule evaluation

Monitor and Iterate

Review violation logs regularly
Test new rules before enabling
Adjust actions based on false positive rates
Use test endpoint before deploying changes

Performance Impact

Guardrails add 50-200ms latency per request
More rules = higher latency
Prioritize critical rules (higher priority number)
Disable unused rules

Gateway API

Management API

Get Guardrail Configuration

Update Guardrail Configuration

Reset Configuration

List Custom Rules

Create Custom Rule

Rule Types

Blocked Terms

Custom Regex

Topic Restriction

Update Custom Rule

Delete Custom Rule

List Violations

Get Violation Statistics

Test Content

List System Rules

Guardrail Actions

Block

Redact

Warn

Allow

Error Responses

Best Practices

Build docs developers (and LLMs) love

Gateway API

Management API

​Get Guardrail Configuration

​Update Guardrail Configuration

​Reset Configuration

​List Custom Rules

​Create Custom Rule

​Rule Types

​Blocked Terms

​Custom Regex

​Topic Restriction

​Update Custom Rule

​Delete Custom Rule

​List Violations

​Get Violation Statistics

​Test Content

​List System Rules

​Guardrail Actions

​Block

​Redact

​Warn

​Allow

​Error Responses

​Best Practices

Build docs developers (and LLMs) love

Get Guardrail Configuration

Update Guardrail Configuration

Reset Configuration

List Custom Rules

Create Custom Rule

Rule Types

Blocked Terms

Custom Regex

Topic Restriction

Update Custom Rule

Delete Custom Rule

List Violations

Get Violation Statistics

Test Content

List System Rules

Guardrail Actions

Block

Redact

Warn

Allow

Error Responses

Best Practices